60220e18b6
Idempotent vLLM OpenAI-compatible serving role, staged-first (does not
start/enable the systemd unit or touch production traffic by default).
Validated end-to-end against astro-orbiter in a brief shadow window
(llama-swap stopped ~5 min, per homelab-llm-inference skill's documented
shadow-validation pattern):
- /health 200, /v1/models returns Qwen2.5-32B-Instruct-AWQ,
/v1/completions live smoke test passes, clean journalctl
- 3 consecutive full-role runs confirmed changed=0 (idempotent)
- production restored: llama-swap active, /v1/embeddings against
nomic-embed-text-v1.5 confirmed still working (Hindsight retain path)
Deviates from the original spec's model choices (Qwen2.5-32B-Instruct /
Qwen3-8B-Instruct bf16) to use the official Qwen AWQ pre-quantized variants
instead -- vLLM does not do safe on-the-fly quantization on this host
(bitsandbytes OOM history) and unquantized bf16 32B does not fit 24GB VRAM.
Two real bugs found+fixed during first-start validation (systemd-only
repro, not visible via interactive SSH testing):
1. ninja not on systemd's minimal PATH -> vLLM torch.compile
FileNotFoundError. Fixed via explicit PATH env in the unit.
2. FlashInfer sampler JIT fails to compile on RTX 3090 (SM86) --
known upstream issue class (vLLM GH #23023, #44305). Fixed via
VLLM_USE_FLASHINFER_SAMPLER=0 (falls back to native sampler).
Also fixed a real idempotency bug: force-upgrading setuptools to latest
fought with vLLM's own setuptools<81.0.0 pin, causing an install/downgrade
flip-flop (changed:true) on every run.
vllm_service_enabled defaults to false -- a host reboot must not
auto-start vLLM and VRAM-collide with the still-live llama-swap production
service. Cutover (enabling + starting + migrating consumers) is an
explicit, separate step outside this role, gated on adding embedding-mode
support (--task embed) for nomic-embed-text-v1.5, which this role does
not yet implement (Hindsight retain still depends on llama-swap's
nomic-embed until that follow-up lands).
Role: roles/deploy-vllm/ (defaults/handlers/meta/tasks/templates/README)
Playbook: playbooks/day1_deploy_vllm.yml
mk-labs
Automated infrastructure provisioning and configuration for a personal homelab, built on GitOps practices with clear tool responsibility boundaries.
Architecture
A single operator action — setting a VM record's status to Staged in NetBox — triggers a fully automated provisioning pipeline:
NetBox (webhook) → n8n (validate & orchestrate) → Terraform (create VM + DHCP)
→ Ansible (OS config + DNS + status update)
| Tool | Host | IP | Responsibility |
|---|---|---|---|
| NetBox | fire-station | 10.1.71.102 | Source of truth — VM records, IP allocation, VLAN data |
| n8n | tiki-room | 10.1.71.23 | Event orchestration, validation, pipeline sequencing |
| Terraform | city-hall | 10.1.71.35 | Proxmox VM lifecycle, Unifi DHCP reservations |
| Ansible / Semaphore | imagineering | 10.1.71.22 | OS configuration, DNS records, NetBox status updates |
| Proxmox | fantasyland | 10.1.71.13 | Target hypervisor |
All systems on the Server Trusted VLAN (10.1.71.0/24).
Repository Structure
homelab/
├── ansible/
│ ├── inventory/ # NetBox dynamic inventory + static
│ ├── playbooks/ # Runnable playbooks (vm-provision, DNS, OS updates)
│ ├── roles/ # vm-baseline, dns-manager, common, haproxy, n8n, observer, etc.
│ ├── tasks/ # Shared includable task files
│ ├── group_vars/ # Group variable definitions
│ ├── host_vars/ # Per-host variable definitions
│ ├── templates/ # Jinja2 templates
│ └── ansible.cfg
│
├── terraform/
│ ├── proxmox/vm/ # bpg/proxmox provider — VM creation from templates
│ ├── unifi/dhcp/ # Unifi provider — DHCP static reservations on UDM Pro
│ └── dns/ # DNS record management
│
├── packer/
│ ├── ubuntu-24.04/ # Ubuntu 24.04 VM template (small → xlarge-plus sizes)
│ └── fedora-42/ # Fedora 42 VM template
│
├── n8n/
│ └── workflows/ # Exported n8n workflow JSON (vm-provisioning)
│
├── netbox/
│ └── initializers/ # Custom fields, VLANs, IP prefixes as code
│
└── docs/
└── decisions/ # Architecture decision records
Pipeline Flow
| # | System | Action |
|---|---|---|
| 1 | NetBox | Operator sets VM status to Staged → webhook fires |
| 2 | n8n | Validates payload (hostname, IP, VLAN, template, proxmox_node) |
| 3 | n8n → city-hall | SSH + terraform apply — creates VM on Proxmox |
| 4 | n8n | Queries Proxmox API for MAC address |
| 5 | n8n → NetBox | Writes MAC to VM interface record |
| 6 | n8n → city-hall | SSH + terraform apply — creates DHCP reservation on UDM Pro |
| 7 | n8n → imagineering | Triggers Ansible via Semaphore API |
| 8 | Ansible | OS baseline, SSH hardening, Technitium DNS A record |
| 9 | Ansible → NetBox | Sets VM status to Active |
On any failure, NetBox status is set to Failed. No auto-retry — operator investigates.
Hardware
- 7× Dell 7050 SFF
- 3× Minisforum TH60
- 2× Minisforum MS01
- Synology DS1621+
- Ubiquiti UDM Pro
Software Stack
- Virtualization: Proxmox
- Automation: Terraform, Ansible, n8n, Semaphore
- DNS: Technitium (authoritative), Unbound (recursive)
- IPAM/DCIM: NetBox
- Networking: Ubiquiti UDM Pro
- Templates: Packer (Ubuntu 24.04, Fedora 42)
Getting Started
See docs/decisions/vm-provisioning-flow.md for the full architecture decision record.
Previous OpenShift/ACM/Fastpass content is preserved in the archive/pre-mk-labs branch.
Security
No sensitive data is stored in this repository. Secrets are managed via Ansible Vault and environment variables on pipeline hosts.
Status: 🚧 Active Development — VM Provisioning Pipeline
Last Updated: February 2026